\documentclass[a4paper, 12pt]{article}
\usepackage{setspace}
\usepackage{fullpage}
\usepackage{amsmath}
\usepackage[protrusion=true,expansion=true]{microtype}
\usepackage{graphicx}
\usepackage{wrapfig}
\usepackage{csquotes}
\usepackage{booktabs}
\usepackage{morefloats}
\usepackage{caption}
\usepackage{threeparttable}
\usepackage{pgf, tikz}
\usetikzlibrary{arrows, automata}
\usepackage{etex}
\usepackage[round]{natbib}
\usepackage{mathpazo}
\usepackage[T1]{fontenc}
\linespread{1.75} 
\usepackage{url}
\usepackage{color}
\usetikzlibrary{positioning,shapes.geometric,arrows,decorations.markings} 
\usepackage{subcaption}
\usepackage[sc]{mathpazo}
\usepackage{hyperref}
\hypersetup{colorlinks=true,linkcolor=black,citecolor=[rgb]{0.2235294, 0.3372549, 0.5098039}} 
\usepackage{tgtermes}
\newtheorem{hypothesis}{Hypothesis}
\usepackage{lscape}
\usepackage{xcolor}
\usepackage{import}
\usepackage{verbatim}
\usepackage{amsmath}
\usepackage{titling}
\usepackage{titletoc}
\usepackage{enumitem}
\usepackage{amssymb}
\usepackage{array}
\usepackage{pdflscape}
\usepackage{afterpage}
\usepackage{siunitx}
\newcolumntype{d}{S[input-symbols = ()]}
\makeatletter
\renewcommand\@biblabel[1]{\textbf{#1.}} 
\renewcommand{\@listI}{\itemsep=0pt} 
\usepackage[symbol]{footmisc}
\makeatletter 
\let\estinput=\@@input 

\newcommand{\estauto}[3]{
	\vspace{.75ex}{
		\begin{tabular}{@{\hskip\tabcolsep\extracolsep\fill}l*{#2}{#3}} 
			\toprule
			\estinput{#1}
			\bottomrule
			\addlinespace[.75ex]
		\end{tabular}
	}
}
		
	\newcommand{\specialcell}[2][c]{%
	\begin{tabular}[#1]{@{}c@{}}#2\end{tabular}}

\usepackage{booktabs}
\usepackage{siunitx}
\newcolumntype{d}{S[input-symbols = ()]}

\newcommand{\Figtext}[1]{%
 \begin{tablenotes}[para,flushleft]
 \hspace{0pt}
 \hangindent=0.25em
 #1
 \end{tablenotes}
}

\newcommand{\StarnoteAllElections}{\Figtext{\footnotesize Heteroskedacity-robust p-values from testing each coefficient equal to 0.5 in brackets. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteDataADC}{\Figtext{\footnotesize \textit{Notes:} Table presents pooled descriptive mean statistics on candidates ranked first ($\mu_1$) and second ($\mu_2$) by ADC. Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). Differences in observations (column 1) are due to variation in source accessibility on candidate-level covariates (see Appendix A.3). $\beta^{ADC}=\mu_1-\mu_2$. Column 5 provides p-value from regressing each candidate covariate onto an indicator for being ranked first with constituency-year fixed effects. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteDataElection}{\Figtext{\footnotesize \textit{Notes:} Table presents pooled descriptive mean statistics on candidates who won election ($\mu_{\text{Winner}}$) and lost ($\mu_{\text{Loser}}$). Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). Differences in observations (column 1) are due to variation in source coverage across candidates. $\beta^{\text{Voters}}=\mu_{\text{Winner}}-\mu_{\text{Loser}}$. Column 4 provides p-value from regressing each candidate covariate onto an indicator for winning election with constituency-year fixed effects. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteDataADCElection}{\Figtext{\scriptsize \textit{Notes:} Panel A presents descriptive statistics on candidates ranked first ($\mu_1$) and second ($\mu_2$) by ADC. $\beta^{ADC}=\mu_1-\mu_2$. Column 5 provides p-value from regressing candidate covariate onto indicator for being ranked first with constituency-year fixed effects. Panel B presents descriptive statistics on candidates who won election ($\mu_{\text{Winner}}$) and lost ($\mu_{\text{Loser}}$). $\beta^{\text{Voters}}=\mu_{\text{Winner}}-\mu_{\text{Loser}}$. Column 9 provides p-value from regressing candidate covariate onto indicator for winning election with constituency-year fixed effects. Column 10 provides p-value from a seemingly unrelated regression to test $\beta^{\text{ADC}}=\beta^{\text{Voters}}$. Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). Differences in observations (column 1) due to variation in source accessibility (see Appendix A.3).  * p < 0.1, ** p < 0.05, *** p < 0.01.}} 

\newcommand{\StarnoteADCWeakStrong}{\Figtext{\footnotesize Table presents differences in mean candidate characteristics between elite-preferred candidates receiving more than the mean ADC vote share ($\mu_1^{\text{Strong}}$) and those elite-preferred candidates receiving less than mean ADC vote share ($\mu_1^{\text{Weak}}$). $\beta$ comes from a regression of a given elite-preferred candidate characteristic on their ADC vote share (vote share conditioned on the top two candidates for comparability) using OLS with election year fixed effects. Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). Column 5 provides p-value associated with $\beta$. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteConstituency}{\Figtext{\footnotesize Table presents differences in mean constituency characteristics across cases where first-ranked candidate won office ($\mu_{\text{Winner}=1}$) and second-ranked candidate won office ($\mu_{\text{Winner}=2}$). Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). $\beta=\mu_{\text{Winner}=1}-\mu_{\text{Winner}=2}$. Column 5 provides p-value of difference using OLS with election year fixed effects. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteBalanceConstit}{\Figtext{\footnotesize Table presents differences in mean predetermined constituency characteristics between constituencies in which elite-preferred candidate was assigned $J$ ($\mu_{1 \rightarrow J}$) and where they were assigned $N$ ($\mu_{1 \rightarrow N}$). Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). $\beta^J=\mu_{1 \rightarrow J}-\mu_{1 \rightarrow N}$. Column 5 provides p-value of difference using OLS with election year fixed effects. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteBalanceCand}{\Figtext{\footnotesize Table presents differences in mean predetermined candidate characteristics (of the elite-preferred candidate) between constituencies in which elite-preferred candidate was assigned $J$ ($\mu_{1 \rightarrow J}$) and where they were assigned $N$ ($\mu_{1 \rightarrow N}$). Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). $\beta^J=\mu_{1 \rightarrow J}-\mu_{1 \rightarrow N}$. Column 5 provides p-value of difference using OLS with election year fixed effects. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteExclusion}{\Figtext{\footnotesize Table presents differences in mean outcomes relating to electoral competitiveness between constituencies in which elite-preferred candidate was assigned $J$ ($\mu_{1 \rightarrow J}$) and where they were assigned $N$ ($\mu_{1 \rightarrow N}$). `Election vote share' refers to vote share of winning candidate. Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). Number of campaign meetings is only selectively reported for 1965 in Cliffe et al (1967). $\beta=\mu_{1 \rightarrow J}-\mu_{1 \rightarrow N}$. Column 5 provides p-value of difference using OLS with election year fixed effects. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteWinners}{\Figtext{\footnotesize Table presents differences in mean characteristics of ultimately elected legislators between constituencies in which elite-preferred candidate was assigned $J$ ($\mu_{1 \rightarrow J}$) and where they were assigned $N$ ($\mu_{1 \rightarrow N}$). Sample restricted to constituencies in 1965/1970 with competitive selection stages (2+ aspirants). $\beta=\mu_{1 \rightarrow J}-\mu_{1 \rightarrow N}$. Column 5 provides p-value of difference using OLS with election year fixed effects. * p < 0.1, ** p < 0.05, *** p < 0.01.}}

\newcommand{\StarnoteFS}{All specifications are estimated using OLS with election-year fixed effects. Even-indexed columns add LASSO-selected controls. DV Mean and SD correspond to constituencies not assigned to instrument. Heteroskedasticity-robust standard errors in parentheses.}

\newcommand{\StarnoteIV}{All specifications are estimated using 2SLS including election year fixed effects. Unit of observation is the constituency-election cycle. Weights based on predicted compliance propensities. DV Mean and SD correspond to constituencies not assigned to instrument. Bootstrapped $p$-values in square brackets.}

\newcommand{\StarnoteIVPoisson}{Specification in Panel A estimated using 2SLS including election year fixed effects; Panel B estimated with Poisson IV model following control function approach of Wooldridge (2010). Unit of observation is the constituency-election cycle. Weights based on predicted compliance propensities. DV Mean and SD correspond to constituencies not assigned to instrument. Bootstrapped $p$-values in square brackets.}

\newcommand{\StarnoteIVDistrict}{All specifications are estimated using 2SLS including election year fixed effects following Equation (A3). Unit of observation is the district-election cycle. DV Mean and SD correspond to districts where no constituencies were assigned the instrument. Estimating sample excludes districts containing any non-competitive ADC selection processes. Bootstrapped $p$-values in square brackets.}

\newcommand{\StarnoteIVSpillovers}{All specifications are estimated using 2SLS including election year fixed effects following Equation (A4). Unit of observation is the constituency-election cycle. DV Mean and SD correspond to districts where no constituencies were assigned the instrument. Estimating sample excludes districts containing any non-competitive ADC selection processes. Bootstrapped $p$-values in square brackets.}


\begin{document}

\begin{center} \large \singlespacing
\textbf{Do Elites Know Best? Candidate Selection and \\ Policy Implementation in Post-independence Tanzania}

\

\normalsize  Jeremy Bowles

\

\textbf{Replication of Tables and Figures}
\end{center}

\addtocontents{toc}{\protect\setcounter{tocdepth}{3}}

\tableofcontents

\clearpage

\section{Main paper}

\setcounter{table}{0}

\afterpage{
\begin{landscape}
\begin{center}
\input{Tables/Table 1.tex}\centering
\end{center}
\end{landscape}
}

\clearpage 

\setcounter{table}{2}

\input{Tables/Table 3.tex}

\begin{table}[!htbp]\centering \small
\begin{threeparttable}
\caption{First stage: Elite-preferred candidate is elected}
\estauto{Tables/Table 4.tex}{10}{c}
\Figtext{\footnotesize Outcome variable: Elite-preferred candidate is elected. \StarnoteFS}
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \small
\begin{threeparttable}
\caption{Effects on supply of local public goods}
\estauto{Tables/Table 5.tex}{15}{c}
\Figtext{\footnotesize Dependent variables: log+1 Number of primary schools/other local public goods founded in constituency in five years following 1965/1970. \StarnoteIV}
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \small
\begin{threeparttable}
\caption{Heterogeneity by ADC vote share of elite-preferred candidate}
\estauto{Tables/Table 6.tex}{15}{c}
\end{threeparttable}
\Figtext{\footnotesize Dependent variables: log+1 Number of primary schools/other local public goods founded in constituency in five years following 1965/1970. ADC vote share measures standardized ADC vote share received by the elite-preferred candidate. \StarnoteIV}
\end{table}

\setcounter{figure}{1}

\begin{figure}[!htpb]
\begin{center}
\caption{Correlation between ADC vote share and election vote share across elections}
\includegraphics[scale=0.9]{./Figures/Figure 2.pdf}
\end{center}
\footnotesize \textit{Notes:} For comparability, ADC vote share conditions on number of ADC votes received by two selected candidates. The same pattern holds when only considering races in which the elite-preferred candidate was not assigned $J$ (see Figure A13).
\end{figure}


\clearpage

\section{Supplementary materials}

\setcounter{table}{0}
\setcounter{figure}{0}
\renewcommand{\thetable}{A\arabic{table}}
\renewcommand{\thefigure}{A\arabic{figure}}

\subsection{Figures}

\begin{figure}[!htpb] \centering
\caption{Global distribution of legislative candidate selection methods}
\includegraphics[scale=0.67]{Figures/Figure A1.pdf} 
\vspace{-0.5cm}
\begin{flushleft}\footnotesize \textit{Notes:} Data comes from V-Party database of political parties (\texttt{v2panom}). V-Party sample includes all parties receiving more than 5\% vote share in national elections (3467 political parties across 178 countries). Right panel restricts to ruling parties.\end{flushleft}
\end{figure}

\begin{figure}[!htpb] 
\begin{center}
\caption{Cross-national scope conditions}
\includegraphics[width=\textwidth]{Figures/Figure A2.pdf} 
\end{center}
\vspace{-0.5cm}
\footnotesize \textit{Notes:} Sample restricted to 1960-1990. Redistributive policy measure comes from V-Party measure of left-wing/redistributive policy of ruling party in a given year (\texttt{v2pariglef}). Regime corruption index comes from V-Dem index of regime corruption (\texttt{v2xnp\_regcorr}), including executive, legislative, and judicial. Presidentialism index comes from V-Dem index of the concentration of political power in one person (\texttt{v2xnp\_pres}), defined as the ``systemic concentration of political power in the hands of one individual who resists delegating all but the most trivial decision making tasks.'' This is used to proxy for the ability of lower-level politicians to influence \textit{some} degree of public resource allocation decisions.
\end{figure}

\begin{figure}[!htpb] \centering
\caption{Sincerity of ADC ranking behavior}
\includegraphics[scale=0.72]{Figures/Figure A3.pdf}
\vspace{-0.5cm}
\begin{flushleft}\footnotesize \textit{Notes:} Figure compares characteristics of selected candidates according to their raw ADC rank. Since NEC held veto rights, in some cases aspirants ranked below 2nd by ADC were ultimately advanced to candidacy.\end{flushleft}
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Constituencies and instrument assignment}
\includegraphics[scale=0.75]{Figures/Figure A4.pdf}
\end{center}
\footnotesize \textit{Notes:} Left panel represents 1965 constituencies; right represents 1970 constituencies. Dark constituencies are assigned instrument, i.e. top-ranked candidate is assigned $J$ ($1 \rightarrow J$); light constituencies are not assigned instrument, i.e. top-ranked candidate is assigned $N$ ($1 \rightarrow N$). Unshaded constituencies have noncompetitive ADC selection stages, tied ADC votes between top two aspirants, or missing ADC voting outcomes in a handful of cases, and hence instrument assignment is undefined.
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Validating administrative data on primary schools}
\includegraphics[width=0.95\textwidth]{Figures/Figure A5.pdf}
\end{center}
\footnotesize \textit{Notes:} X-axis records the number of primary schools existing in each district as reported by government sources in either 1967 or 1973. Y-axis records the number of primary schools, based on the facility-level administrative data where I observe year of foundation, which still exist and were founded prior to either year.
1967 source: Jensen \& Mkama (1968); 1973 source: \textit{Hansard}, 25-30 June 1973, pp. 1009-1111. 
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Provision of different local public goods by year, 1960-1980}
\includegraphics[scale=0.55]{Figures/Figure A6.pdf}
\end{center}
\footnotesize \textit{Notes:} Number of local public goods (by type) founded by year between 1960 and 1980 as observed in administrative datasets. 
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Distribution of outcome measures}
\includegraphics[scale=0.6]{Figures/Figure A7.pdf}
\end{center}
\footnotesize \textit{Notes:} Histograms of the number of local public goods (by type) aggregated to the constituency level founded in the five years following either the 1965 or 1970 election. `Other local public goods' consists of dispensaries, other health facilities (primarily hospitals), secondary schools, and water points (bottom row). 
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Spatial distribution of outcome measures}
\includegraphics[scale=0.6]{Figures/Figure A8.pdf}
\end{center}
\footnotesize \textit{Notes:} Outcomes relating to supply of primary schools and other public goods provided. Left panel represents 1965 constituencies; right represents 1970 constituencies. Unshaded constituencies have noncompetitive ADC selection stages, tied ADC votes between top two aspirants, or missing ADC voting outcomes in a handful of cases, and hence instrument assignment is undefined (and so these constituencies are excluded from the analysis sample. 
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Estimates of first stage while excluding districts or regions}
\includegraphics[scale=0.75]{Figures/Figure A9.pdf}
\end{center}
\footnotesize \textit{Notes:} Figure plots $\beta^{FS}$ from Equation (1) while sequentially excluding each district-year (left) or region-year (right). Coefficients ordered by magnitude. 95\% confidence intervals plotted. 
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Heterogeneity in estimated first stage effect}
\includegraphics[scale=0.6]{Figures/Figure A10.pdf}
\end{center}
\footnotesize \textit{Notes:} Figure plots the distribution of predicted first stage treatment effects obtained through a causal forest (Wager \& Athey, 2018). Estimating using cross-fitting by obtaining split-sample estimates from 1000 splits and taking the median across splits. 
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Correlates of first stage effect heterogeneity}
\includegraphics[scale=0.7]{Figures/Figure A11.pdf}
\end{center} \vspace{-0.5cm}
\footnotesize \textit{Notes:} Figure plots the 20 variables with the largest absolute coefficients arising from a regression of the predicted first stage treatment effect (based on causal forest) on that variable.
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Baseline estimates from Table 5 while excluding districts or regions}
\includegraphics[scale=0.75]{Figures/Figure A12.pdf}
\end{center}
\footnotesize \textit{Notes:} Figure plots $\beta^{IV}$ from Equation (2) while sequentially excluding each district-year (top) or region-year (bottom). Coefficients ordered by magnitude. 95\% confidence sets from wild bootstrap permutations plotted.
\end{figure}

\begin{figure}[!htpb]
\begin{center}
\caption{Correlation between ADC vote share and election vote share across elections (in constituencies where elite-preferred candidate is not assigned $J$)}
\includegraphics[scale=0.75]{Figures/Figure A13.pdf}
\end{center}
\footnotesize \textit{Notes:} For comparability, ADC vote share conditions only on the number of ADC votes received by the two candidates ultimately running in the election.
\end{figure}

\begin{center}
\begin{figure}[!htpb]
\caption{Central government capital allocations across public goods}
\begin{center}
\includegraphics[scale=0.65]{Figures/Figure A14.pdf} \end{center} \\ 
\footnotesize \textit{Notes}: Measures constructed using programme-level data from Vols. II of \textit{First Five-Year Plan (1964-69) and \textit{Second Five-Year Plan (1969-74)}.}
\end{figure}
\end{center}

\begin{figure}[!htpb] 
\begin{center}
\caption{Distribution of distance to facilities in HRDS data (1993)}
\includegraphics[width=\textwidth]{Figures/Figure A15.pdf}
\end{center}
\vspace{-0.5cm}
\footnotesize \textit{Notes:} Figure provides the distance between each household in the HRDS sample and its closest facility. Averages for each type of facility are provided with the blue dotted lines. X-axis is log-transformed.
\end{figure}


\clearpage

\subsection{Tables}

\begin{landscape}\input{Tables/Table A1.tex}\end{landscape}

\begin{landscape}\input{Tables/Table A2.tex}\end{landscape}

\begin{landscape}\input{Tables/Table A3.tex}\end{landscape}

\input{Tables/Table A4.tex}

\begin{landscape}\input{Tables/Table A5.tex}\end{landscape}

\begin{landscape}\input{Tables/Table A6.tex}\end{landscape}

\input{Tables/Table A7.tex}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Elite-preferred candidate is elected (by election)}
\estauto{Tables/Table A8.tex}{10}{c}
\Figtext{\scriptsize DV: Elite-preferred candidate is elected. Table A25 displays LASSO-selected control coefficients. \StarnoteFS}
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Disaggregating effects on supply of other local public goods}
\estauto{Tables/Table A9.tex}{15}{c}
\Figtext{\scriptsize Dependent variables: log+1 Number of dispensaries; other health facilities; secondary schools; water points, founded in a given constituency in the five years following 1965 or 1970 elections. \StarnoteIV} 
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Effects on supply of local public goods (Sample exclusions)}
\estauto{Tables/Table A10.tex}{15}{c} 
\Figtext{\scriptsize Dependent variables: log+1 Number of primary schools/other local public goods founded in a given constituency in the five years following 1965 or 1970 elections. Panel A includes facilities in wards in major towns (excluded in baseline estimation). Panel B excludes all constituencies in which National Executive Committee exercised its veto over the aspirant ranked first by ADC (and hence \textit{elite-preferred} candidate had ranked second or lower in ADC). \StarnoteIV}
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Effects on supply of local public goods (Varying outcomes)}
\estauto{Tables/Table A11.tex}{15}{c}
\Figtext{\scriptsize Dependent variables: Panel A: Inverse hyperbolic sine (IHS) transformation of Number of primary schools/other local public goods founded in a given constituency in the five years following 1965 or 1970 elections. Panel B: Non-transformed count of the same outcomes using a Poisson model second-stage. Panel C: Log per 1000 population (based on 1957 census) measure of the same outcomes. \StarnoteIVPoisson}  
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Effects on distance to closest facilities (HRDS)}
\estauto{Tables/Table A12.tex}{12}{c}
\Figtext{\scriptsize \textit{Notes:} Data source is the \textit{Human Resource Development Survey} (1993). Dependent variables: Log-transformed distance in meters between respondent household and closest primary school (Panel A); secondary school (Panel B); dispensary (Panel C); health facility (Panel D); water point (Panel E). All specifications are estimated using 2SLS including election year fixed effects and rural cluster fixed effect. Unit of observation is the household-election cycle. Weights based on predicted compliance propensities. DV Mean and SD correspond to households in constituencies not assigned to instrument. Standard errors clustered by constituency-election cycle. Bootstrapped $p$-values in square brackets.}  
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \footnotesize
\begin{threeparttable}
\caption{Effects on reported quality of closest primary school (HRDS)}
\estauto{Tables/Table A13.tex}{12}{c}
\Figtext{\scriptsize \textit{Notes:} Data source is the \textit{Human Resource Development Survey} (1993). Dependent variables: Aggregated index of questions about household head's perception of quality of closest primary school; comprising teachers' quality, headteachers' quality, school supplies, high quality facilities, student learning outcomes in literacy and numeracy. Columns 1-4 aggregate these using a standardized z-score; columns 5-8 aggregate these using the standardized first principal component. All specifications are estimated using 2SLS including election year fixed effects and rural cluster fixed effect. Unit of observation is the household-election cycle. Weights based on predicted compliance propensities. DV Mean and SD correspond to households in constituencies not assigned to instrument. Standard errors clustered by constituency-election cycle. Bootstrapped $p$-values in square brackets.} 
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \footnotesize
\begin{threeparttable}
\caption{Effects on educational attainment (HRDS)}
\estauto{Tables/Table A14.tex}{12}{c}
\Figtext{\scriptsize \textit{Notes:} Data source is the \textit{Human Resource Development Survey} (1993). Dependent variables: Respondent has any formal education (columns 1-4); respondent can read and write (columns 5-8). Unit of observation is the individual-election cycle. All specifications are estimated using 2SLS including fixed effects for election cycle, rural cluster, year of birth, and gender. Panel A restricts to cohorts who were seven years old at any point during a given electoral cycle; panel B extends to the five cohorts beyond this. Weights based on predicted compliance propensities. DV Mean and SD correspond to individuals in constituencies not assigned to instrument. Standard errors clustered by constituency-election cycle. Bootstrapped $p$-values in square brackets.} 
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Effects on supply of local public goods (district-level)}
\estauto{Tables/Table A15.tex}{15}{c}
\Figtext{\scriptsize Dependent variables: log+1 Number of primary schools/other local public goods founded in a given district in the five years following 1965 or 1970 elections. \StarnoteIVDistrict}  
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Effects on supply of local public goods (spillovers test)}
\estauto{Tables/Table A16.tex}{10}{c}
\Figtext{\scriptsize Dependent variables: log+1 Number of primary schools/other local public goods founded in a given constituency in the five years following 1965 or 1970 elections. \StarnoteIVSpillovers}  
\end{threeparttable}
\end{table}

\begin{table}[!htbp] \centering \scriptsize
\begin{threeparttable}
\caption{Effects on local government employment}
\estauto{Tables/Table A17.tex}{15}{c}
\Figtext{\scriptsize Dependent variables: Any new government employees are observed within 1-4 quarters of election in local government offices within constituency; Any new postings (i.e. existing bureaucrats given new jobs) are observed within 1-4 quarters of election in local government offices within constituency. Panel A: Outcomes are indicators for any new employees/postings; Panel B: Outcomes are log+1 number of new government employees/number of new postings; Panel C: Excluding constituencies not observed as containing a local government office at any point; Panel D: Excluding Dar es Salaam; Panel E: Excluding cases where elite-preferred candidate was incumbent. \StarnoteIV}
\end{threeparttable}
\end{table}

\input{Tables/Table A18.tex}

\input{Tables/Table A19.tex}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{First stage heterogeneity by ADC vote share of elite-preferred candidate}
\estauto{Tables/Table A20.tex}{10}{c}
\Figtext{\scriptsize DV: Elite-preferred candidate is elected. ADC vote share measures the standardized ADC vote share received by the elite-preferred candidate during the selection stage. \StarnoteFS}
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Other local public goods: Heterogeneity by ADC vote share of elite-preferred candidate}
\estauto{Tables/Table A21.tex}{15}{c}
\Figtext{\scriptsize Dependent variables: log+1 Number of dispensaries; other health facilities; secondary schools; water points, founded in a given constituency in the five years following 1965 or 1970 elections. \StarnoteIV} 
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Heterogeneity by ADC vote share of elite-preferred candidate (by tercile)}
\estauto{Tables/Table A22.tex}{15}{c}
\Figtext{\scriptsize Dependent variables: log+1 Number of primary schools/other local public goods founded in a given constituency in the five years following 1965 or 1970 elections. ADC vote share measures the tercile of the ADC vote share received by the elite-preferred candidate during the selection stage. \StarnoteIV} 
\end{threeparttable}
\end{table}

\begin{table}[!htbp]\centering \scriptsize
\begin{threeparttable}
\caption{Effects on supply of local public goods (by election)}
\estauto{Tables/Table A23.tex}{15}{c}
\Figtext{\scriptsize Dependent variables: log+1 Number of primary schools/other local public goods founded in a given district in the five years following 1965 or 1970 elections. \StarnoteIV} 
\end{threeparttable}
\end{table}

\end{document}